Research Of ECG Identification Technology Based On Convolutional Neural Network

With the rapid spread of network and the rapid development of information technology,personal identification based on biometric technology has been widely used in many fields such as data security,medical treatment and financial security.Compared with the traditional biometric method,ECG identification has an outstanding characteristic of anti-spoof and liveness detection.In recent years,the emergence of the ECG acquisition chips with small volume,low energy consumption,avoiding the usage of conductive glue and easy integration represents a huge potential and bright prospect for the development and application of ECG identification.Based on the ECG signals collected by portable mobile devices,this paper has carried out in-depth analysis and research from these three aspects:ECG signal quality assessment,preprocessing and identification.Firstly,in order to eliminate the ECG signals with unsatisfactory acquisition and poor waveform performance,an ECG quality assessment mechanism based on simple heuristic fusion and fuzzy comprehensive evaluation is proposed.Secondly,different pre-processing schemes are adopted for different quality ECG signals.Finally,a LeNet-5-based convolutional neural network ECG authentication system is used to effectively identify individual identities.A total of 750 sets of ECG signals from 6 databases of different signal behaviors in PhysioBank were used as data sources.10-fold cross-validation methods were used 10 times to optimize and evaluate algorithms and models with six performance indicators such as accuracy and error rate.Summarize the main work and innovation features of this paper are as follows:Summarize the main work and features of this paper are as follows:(1)Systematically expounded the theoretical basis of ECG signal processing,including the generation mechanism,acquisition mode,waveform characteristics and noise characteristics,as well as the database and model training methods used in model training,which provides theoretical basis for subsequent research work.(2)A new ECG quality assessment system is proposed:SQI quality evaluation mechanism of ECG signal based on simple heuristic fusion and fuzzy comprehensive evaluation.Firstly,six SQI signal quality indices are introduced to quantify ECG quality.Then,only four SQI parameters(qSQI,pSQI,kSQI,basSQI)are extracted by simple heuristic fusion technology to fully reflect the quality of ECG signals,avoiding the computational complexity and high redundancy of the traditional multiple SQI indices.Finally,fuzzy comprehensive evaluation is used to achieve accurate measurement.The performance of the proposed method was tested on the database from Physionet/Cinc Challenge 2011 and Physionet/Cinc Challenge 2017,within 600 data sets,carried out 10 times of 10-fold cross-validation method,94.67%accuracy evaluation was obtained.(3)A pre-processing scheme with full frequency domain characterization and without peak detection is designed:corresponding noise cancellation processing is performed according to the result of quality assessment.The blind segmentation technique is applied to signal segments to efectively avoid the complexities of waveform recognition and segmentation techniques of ECG signals,such as R-peak detection and QRS wave identification.Then,the generalized S-transformation is introduced,combined with the getframe technology,to convert the signal from the time domain to the frequency domain and achieve the conversion of one-dimensional signals to two-dimensional images.The ECG trajectory map(a total of m ECG trajectory maps at m time points)can fully reflect the change trend of the ECG signal spectrum characteristics in the continuous period,that is,the quasi-periodicity and uniqueness of the ECG signal in the time domain are reflected in the frequency domain.(4)An ECG authentication system with high precision and strong robustness is constructed:LeNet-5-based convolutional neural network ECG authentication system.The convolutional neural network(CNN)is introduced,with all of the ECG trajectory maps(in the frequency domain)as the input layer of the CNN.The structure of the convolutional neural network is optimized,through the feature self-learning,the workload of feature engineering is effectively reduced,and the intrinsic feature patterns is captured more effectively.Considering the possible impact of ECG signals with different signal behaviors on identity recognition,we used 150 groups of normal individual,AF patient,and noisy databases from Physionet/Cinc Challenge 2017 and ECG-ID Database for performance assessment,achieved 96.63%,96.23%and 96.18%accurate identification,respectively.The research work in this paper has achieved a high-precision quality assessment and identification of ECG signals,providing a new way of biometric identification with strong robustness and high anti-counterfeiting ability for tele-medicine,home health care and information security,providing the theoretical basis and technical support for the large-scale application of ECG identification technology.